ProseMirror's support for collaborative editing requires an "authority server". From the ProseMirror documentation:
The role of the central authority is actually rather simple. It must...
- Track a current document version
- Accept changes from editors, and when these can be applied, add them to its list of changes
- Provide a way for editors to receive changes since a given version
The "track a current document version" and "Provide a way for editors to receive changes" aspects of what an Authority Server does are relatively runtime agnostic — a server (or even someone else's API), written in any runtime, can perform these without any specific knowledge of ProseMirror.
The "Accept changes from editors" part does require some component to be ProseMirror-aware. ProseMirror's libraries are written in Javascript, targeting the browser and node.js. The reference implementation of an authority server stores the current document and steps in-memory in a node.js server, and uses functions exported by ProseMirror libraries to apply steps to a document. This is probably how many authority server implementations work.
This repo tries to evaluate alternatives to running an authority server in node. There may be many reasons to do this — a service already written in an existing runtime, performance considerations surrounding node vs. other language runtimes — that are beyond the scope of this example.
This repo has a node.js and Go implementation of an HTTP endpoint that receives a ProseMirror document and array of steps, hydrates them using a specific ProseMirror schema, applies the steps to the document, and returns the results. The benchmark scripts start each server, and send requests to them, collecting latency information.
Requires: go, node, and k6 (for load testing).
The node.js implementation imports ProseMirror libraries directly, and is quite simple.
Test using:
./benchmark-node.shThe Go implementation configures and runs the ProseMirror libraries indirectly, using the v8go package, which allows Go to create and run v8 isolates (which are tiny Javascript VM's).
Test using:
./benchmark-go.shNode:
scenarios: (100.00%) 1 scenario, 40 max VUs, 1m0s max duration (incl. graceful stop):
* default: 40 looping VUs for 30s (gracefulStop: 30s)
data_received..................: 352 kB 12 kB/s
data_sent......................: 311 kB 10 kB/s
http_req_blocked...............: avg=70.88µs min=0s med=4µs max=2.31ms p(90)=7.1µs p(95)=11.04µs
http_req_connecting............: avg=38.79µs min=0s med=0s max=1.33ms p(90)=0s p(95)=0s
http_req_duration..............: avg=3.45ms min=344µs med=3.08ms max=16.55ms p(90)=5.25ms p(95)=6.22ms
{ expected_response:true }...: avg=3.45ms min=344µs med=3.08ms max=16.55ms p(90)=5.25ms p(95)=6.22ms
http_req_failed................: 0.00% ✓ 0 ✗ 1200
http_req_receiving.............: avg=28.4µs min=7µs med=25µs max=287µs p(90)=45µs p(95)=56µs
http_req_sending...............: avg=17.19µs min=4µs med=14µs max=174µs p(90)=31µs p(95)=40.04µs
http_req_tls_handshaking.......: avg=0s min=0s med=0s max=0s p(90)=0s p(95)=0s
http_req_waiting...............: avg=3.4ms min=325µs med=3.03ms max=16.5ms p(90)=5.21ms p(95)=6.2ms
http_reqs......................: 1200 39.840717/s
iteration_duration.............: avg=1s min=1s med=1s max=1.01s p(90)=1s p(95)=1s
iterations.....................: 1200 39.840717/s
vus............................: 40 min=40 max=40
vus_max........................: 40 min=40 max=40
Go:
scenarios: (100.00%) 1 scenario, 40 max VUs, 1m0s max duration (incl. graceful stop):
* default: 40 looping VUs for 30s (gracefulStop: 30s)
data_received..................: 220 kB 7.3 kB/s
data_sent......................: 311 kB 10 kB/s
http_req_blocked...............: avg=71µs min=1µs med=4µs max=2.34ms p(90)=11µs p(95)=16µs
http_req_connecting............: avg=37.26µs min=0s med=0s max=1.46ms p(90)=0s p(95)=0s
http_req_duration..............: avg=2.64ms min=199µs med=2.38ms max=13.39ms p(90)=5.04ms p(95)=6.03ms
{ expected_response:true }...: avg=2.64ms min=199µs med=2.38ms max=13.39ms p(90)=5.04ms p(95)=6.03ms
http_req_failed................: 0.00% ✓ 0 ✗ 1200
http_req_receiving.............: avg=40.38µs min=4µs med=26µs max=4.12ms p(90)=49µs p(95)=82.04µs
http_req_sending...............: avg=54.41µs min=6µs med=16µs max=6.05ms p(90)=85µs p(95)=160.14µs
http_req_tls_handshaking.......: avg=0s min=0s med=0s max=0s p(90)=0s p(95)=0s
http_req_waiting...............: avg=2.55ms min=174µs med=2.29ms max=11.86ms p(90)=4.88ms p(95)=5.84ms
http_reqs......................: 1200 39.871942/s
iteration_duration.............: avg=1s min=1s med=1s max=1.01s p(90)=1s p(95)=1s
iterations.....................: 1200 39.871942/s
vus............................: 40 min=40 max=40
vus_max........................: 40 min=40 max=40